SFormer: SNR-guided Transformer for Underwater Image Enhancement from the Frequency Domain

2025-08-26

AI Summary

Existing underwater image enhancement methods struggle to disentangle cross-channel interference when leveraging signal-to-noise ratio (SNR) priors to suppress wavelength-dependent attenuation, and simultaneously balance structural enhancement with noise suppression. To address these challenges, this paper proposes a frequency-domain SNR-guided Transformer model. We introduce the SNR prior into the frequency domain for the first time, designing Fourier attention (SNR-aware) and frequency-adaptive gating mechanisms to achieve amplitude-phase decoupling, cross-channel interference separation, and joint optimization of low- and high-frequency components. A U-shaped dual-stream architecture is adopted to improve feature reuse efficiency. Trained on 4,800 image pairs, our method achieves a 3.1 dB PSNR gain and a 0.08 SSIM improvement over state-of-the-art approaches. It significantly restores color fidelity, texture details, and global contrast. This work establishes an interpretable, frequency-domain-driven paradigm for underwater visual enhancement.

Technology Category

• Computer Vision: Multi-modal Vision
• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Recent learning-based underwater image enhancement (UIE) methods have advanced by incorporating physical priors into deep neural networks, particularly using the signal-to-noise ratio (SNR) prior to reduce wavelength-dependent attenuation. However, spatial domain SNR priors have two limitations: (i) they cannot effectively separate cross-channel interference, and (ii) they provide limited help in amplifying informative structures while suppressing noise. To overcome these, we propose using the SNR prior in the frequency domain, decomposing features into amplitude and phase spectra for better channel modulation. We introduce the Fourier Attention SNR-prior Transformer (FAST), combining spectral interactions with SNR cues to highlight key spectral components. Additionally, the Frequency Adaptive Transformer (FAT) bottleneck merges low- and high-frequency branches using a gated attention mechanism to enhance perceptual quality. Embedded in a unified U-shaped architecture, these modules integrate a conventional RGB stream with an SNR-guided branch, forming SFormer. Trained on 4,800 paired images from UIEB, EUVP, and LSUI, SFormer surpasses recent methods with a 3.1 dB gain in PSNR and 0.08 in SSIM, successfully restoring colors, textures, and contrast in underwater scenes.

Problem

• Addressing spatial domain SNR prior limitations in underwater image enhancement
• Separating cross-channel interference and amplifying informative structures effectively
• Restoring colors, textures, and contrast in degraded underwater images

Innovation

• SNR prior in frequency domain
• Fourier Attention SNR-prior Transformer
• Frequency Adaptive Transformer bottleneck